Statistical extreme value modelling to study roadside air pollution episodes

• Main Author: Gyarmati-Szabo, Janos
• Published: University of Leeds 2011
• Subjects: 363.7392
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.551267

Motivated by the potential danger of high air pollution concentrations (episodes) on human health and the environment, the overall aim of this thesis is to gain a greater understanding of and insight into the formation of such episodic conditions via proposing new extreme value statistical models. The modelling and prediction of air pollution episodes' occurrence, strength and dur~tion are formidable problems in the urban atmospheric media due to the combination of many complex simultaneously working physical and chemical processes involved in their formations. It has been long observed that conventional statistical methods may not be suitable for solving these problems, thus initiating the application of more flexible approaches. In the last couple of decades Extreme Value Theory (EVT) has been widely used with great success to overcome some of the aforementioned issues. However, even the most recent EVT models cannot deal with all the aspects of these problems. The objective of this research is to specify the requirements of new extreme value models by taking into account the demerits of the old ones, to develop such new models and validate their adequacy on real datasets. To place this research in relation to the wide-ranging existing literature and to identify the model requirements, a comprehensive review on EVT and its applications in air pollution modelling has been conducted. Based on the gaps identified in the literature, four extreme value models are proposed in the Peaks over Threshold context, which are either improvements on existing models or completely new ones involving new theoretical results in the background. Based on these models, and their possible amalgamations, the occurrence times, the strengths and the durations of episodes can be modelled and predicted. The relationship between these characteristics and meteorological as well as traffic conditions are identified, which are considered as the most significant contributors to these events.